Our main research topic is multilingual Natural Language Processing (NLP). While many language-independent NLP techniques are useful, such techniques often fail to capture differences on various granularity among languages. By capturing the differences with using multilingual text in the Web, we aim at contributing improvement of cross-lingual NLP techniques such as machine translation.
Automatic linking to Wikipedia articles using multilinguality of Wikipedia
A Wikipedia article has internal links for connecting words in the text to articles describing about them. These links are important for helping readers understand the article. However, annotating the links is a time-consuming task for Wikipedia editors. Recently, wikification (i.e., automatic linking) is an attracting task for NLP-related researchers. We study on wikification with using multilingual information of Wikipedia.
Improving usability on multilingual use of Wikipedia
Wikipedia articles in each language version are not necessarily consistent, that is, different language versions may have different contents, because they were edited independently. We are developing a system for displaying correspondences of text in different language versions, which facilitate understanding of other language versions. Another topic is building a language-independent knowledge structure from diverse Wikipedia category structures of each language.
Own web site
Cross-language transfer of internal links in Wikipedia
Building Wikipedia multilingual category structure